Is It Too Late to Buy Nvidia and Broadcom? Here's What History Tells Us

Nvidia reported 73% growth in the latest quarter and expects 77% growth next quarter, while Broadcom’s AI semiconductor revenue rose 106% to $8.4 billion. Broadcom said its custom AI chips business could reach $100 billion or more in annual revenue by the end of 2027. The article argues both stocks remain attractively valued on forward earnings despite strong year-to-date gains of 36% for Nvidia and over 50% for Broadcom.

Analysis

The market is implicitly debating two different moats, and that distinction matters more than the headline growth rates. NVDA still owns the software-plus-platform stack, which makes it the cleaner beneficiary when AI deployments stay heterogeneous and models keep changing; AVGO benefits when workloads harden and hyperscalers decide the economics of customization outweigh flexibility. That means the competitive battleground is not share theft between the two so much as the mix shift between generalized training/inference spend and highly specific, cost-optimized deployment spend.

Second-order beneficiaries are the ecosystem enablers, not the obvious semiconductor peers. If custom silicon adoption keeps inflecting, the real lever is in design wins, networking, and packaging capacity, which should keep pressure on supply-chain bottlenecks and preserve pricing power for a narrow set of vendors. The main loser is not Intel as a whole, but any incumbent merchant silicon supplier whose roadmap depends on AI customers tolerating inferior total cost of ownership; that creates a longer-duration competitive threat to legacy datacenter refresh cycles.

The risk is that consensus is extrapolating unit growth into a straight-line revenue runway without pricing in digestion. AI capex can stay strong while ordering becomes more lumpy, and both names are vulnerable to a 1-2 quarter pause if hyperscalers digest prior deployments or shift mix toward internally designed chips faster than expected. In that scenario, the stocks can derate before the earnings estimates do, especially if forward multiples are already anchored to perfection.

Contrarian take: the market may be underestimating how durable the ‘picks-and-shovels’ layer remains even if the model layer commoditizes. If inference cost curves keep falling, demand can widen faster than unit economics compress, which supports both NVDA and AVGO rather than forcing a winner-take-most outcome. The setup argues for owning both, but sizing AVGO as the slower, more leveraged second-derivative trade and NVDA as the higher-quality core compounder.

Key Decisions for Investors

• Overweight NVDA on any 3-5% pullback over the next 2-4 weeks; best risk/reward as the core AI platform winner with less dependence on a single customer design cycle.
• Add AVGO as a paired second-order AI trade over the next 1-3 months, but size smaller than NVDA; upside is greater if custom silicon adoption accelerates, downside is a slower burn if hyperscaler capex pauses.
• Avoid chasing INTC strength into results; use any AI-driven sympathy rally as an opportunity to fade, since custom AI and platform share gains are structurally more likely to pressure legacy datacenter upgrade economics.
• For tactical expression, consider a long NVDA / short INTC pair for 1-2 quarters to isolate AI platform share capture versus legacy erosion; thesis works if spending stays high but customer preference keeps shifting to best-in-class architectures.
• If you want convexity, buy NVDA call spreads 3-6 months out on dips rather than outright stock; this captures continuation of accelerating growth while limiting damage if hyperscaler digestion causes a temporary multiple reset.